EN 18286 clause 8 and clauses 9.1–9.5: lifecycle and operations

EN 18286:2026 is a voluntary European Standard for a quality management system (QMS) for EU AI Act purposes. This page gives the requirements of clause 8 (AI system realization) and of clauses 9.1 to 9.5 of clause 9 (Operations and control). In EN 18286, "shall" gives a requirement of the standard, and "should" shows that the standard only recommends an action (clause 0.2).

Clauses 4 to 7 and 10 are on The QMS clauses, and clauses 9.6 (serious incidents) and 9.7 (non-compliance) are on Incidents and non-compliance. Modulos holds clause 8 and clauses 9.1 to 9.5 in the AI System Framework MFF-19 (see Clause 8 and clauses 9.1–9.5 in Modulos).

Clause 8 — AI system realization

EN 18286 clause 8 gives the requirements for the processes of the provider for each AI system, from inception to retirement.

8.1 Processes and procedures. EN 18286 requires the provider to establish processes and procedures for the applicable life cycle stages of the AI system (8.1.1). The provider also documents, implements and maintains them. These processes and procedures:

• Make sure that all applicable regulatory requirements are fulfilled, and stay fulfilled.
• Are "appropriate" to make sure that the AI system requirements are fulfilled, and stay fulfilled.
• Are necessary to do the actions that the provider determined in clause 6.

As applicable, the processes and procedures include the methods for inception, design and development, verification and validation, operation and monitoring, modifications, and retirement (8.1.3). They also include the methods for data management, documentation, identification of the AI system, supply chain management, post-market monitoring, serious-incident reporting, and non-compliance.

For these processes, the provider sets criteria and determines their sequence and interaction (8.1.4). The provider also determines the methods and criteria that make the operation and the supervision of the processes effective. The provider examines the items that 8.1.4 d) gives. Examples are the AI system requirements, the necessary verification and validation, and the responsibilities from 5.3. If applicable, the provider also examines if it is necessary to include "relevant" interested parties, for example deployers and affected persons, in "relevant" processes. The provider monitors the effectiveness of the processes and procedures, and does corrective actions if they do not give the results that they are for.

Documented information is available to the level that is necessary to show that the processes agreed with the plans (8.1.5). For each AI system, the provider maintains the documents for planning and process control, and keeps them up to date as the life cycle continues. A NOTE to 8.1.2 tells providers that the life cycle stages do not always occur in a linear sequence. The provider can do a stage again.

8.2 Risk management system. EN 18286 requires the provider to have a documented risk management system for each AI system (8.2). The provider establishes the system, implements it, and maintains it for the full life cycle of the AI system, as the regulatory requirements require. The system is for a high level of protection of health, safety and fundamental rights. A NOTE to 8.2 refers to EN 18228 for more information about risk management. When EN 18286 was published, EN 18228 was under preparation, at the stage prEN 18228:2026 (Bibliography [6]).

In Table ZA.1 of EN 18286, the rows for Article 17(1)(b), (c), (d) and (g) of the EU AI Act include clause 8.2. Only the row for point (g) has a condition: the risk management system complies with Article 9. Annex ZA connects these rows to a presumption of conformity. This presumption applies from the citation of EN 18286 in the Official Journal on 14 October 2026, by Implementing Decision (EU) 2026/2274 (see Annex ZA and presumption).

8.3 Inception, design and development. At inception, EN 18286 requires the provider to determine the intended purpose for the AI system (8.3.1.1). The standard recommends that the provider examine if consultation with affected persons, or with persons who speak for them, is necessary. The consultation is about possible risks to health, safety and fundamental rights, as the risk assessment shows (8.3.1.2 and Annex A).

For design and development, the provider determines the AI system requirements for the intended purpose (8.3.2.1.1). These requirements are necessary for compliance with the applicable regulatory requirements, which include the essential requirements. EN 18286 requires that the AI system requirements are complete and clear, and that the provider can verify and validate them (8.3.2.1.2).

The provider does a review of the AI system requirements for adequacy, and gives approval to them (8.3.2.2). The review is systematic. It occurs before the provider writes the AI system specifications, and at regular intervals during design, development, verification and validation. The results of the review, and the actions from it, are available as documented information.

The AI system specifications tell how the provider gives the AI system its design and implements it to fulfill the AI system requirements (8.3.2.3). EN 18286 requires that the specifications agree with all of these conditions:

• They fulfill the AI system requirements.
• They give information about the AI system and its design, and about the items that are "relevant" to fulfill the regulatory requirements.
• They are verifiable.

The provider keeps the specifications in the technical documentation (8.9.2).

For design and development, EN 18286 requires the provider to make sure of these items (8.3.2.4):

• Verification, validation, and design control (for example reviews) show that the design and development specifications fulfill the AI system requirements.
• The provider does the necessary actions for the problems that verification and validation find.
• The provider keeps documented information about the verification, the validation, and the design control.

8.4 Verification and validation. EN 18286 makes verification and validation a part of quality control and quality assurance (8.4.1). The two have different requirements.

Verification, which includes tests, makes sure that each AI system fulfills its AI system specifications (8.4.2). EN 18286 requires the provider to write and document testing plans and test procedures. They agree with all of these conditions:

• They are "appropriate" for the intended purpose and for the reasonably foreseeable misuse that the provider identified.
• They give the methods and the criteria to accept test results: limits in numbers, ranges, or other "suitable" measures that are verifiable.
• They agree with "best practices" for the AI technologies, and they are reproducible.

For reproducibility, the provider documents the conditions for testing (8.4.2 c)). As applicable, these conditions include:

• The hardware and compute environment
• The software frameworks, programming languages and library versions
• The random seeds and initialization procedures
• The training parameters and hyperparameters of the model
• The dataset versions and preprocessing steps
• Each stochastic part, and the method that the provider uses for it

A NOTE to 8.4.2 is about stochastic AI systems. For these AI systems, reproducibility can include documented statistical limits for the possible differences in the results when the provider does the same test again. These limits are an alternative to replication with the same results.

EN 18286 requires the provider to maintain documented information about the results and the "conclusions" of the verification (8.4.2 and 4.5).

Validation makes sure that each AI system can fulfill the requirements for its intended purpose (8.4.3.1). EN 18286 requires that validation agrees with all of these conditions:

• It follows documented procedures, for example the instructions for use.
• It uses the instructions for use and the technical documentation of the AI system.
• It occurs during the development of each AI system, and after it.
• It is complete before placing on the market or putting into service.
• It includes documented validation plans and test procedures. These give the methods and the criteria to accept test results: limits in numbers, ranges, or other "suitable" measures.

Clause 8.4.3.1 does not add the requirement of 8.4.2 for "best practices" and reproducibility. The provider determines how frequently it does validation. The provider also does an evaluation of the residual risk from the validation results. The provider keeps the results, the "conclusion", and the necessary actions as documented information. Clause 8.4.3.2, with Annex A, recommends that the provider examine if consultation with interested parties about possible risks to health, safety and fundamental rights is necessary.

8.5 Data management. In EN 18286, data includes, as applicable, training, validation, testing, logging and input data (8.5). It also includes other data for placing on the market or putting into service. EN 18286 requires the provider to have a strategy for compliance with the applicable regulatory requirements on data management. The strategy agrees with clause 4.4.

For each AI system, the provider sets, documents and implements data management processes for design, development, verification and validation. As "appropriate", and in proportion to the risk of each AI system, the provider does these tasks:

• It sets and documents the data requirements.
• It establishes and maintains systems and procedures for data management.

These systems and procedures include the items in clause 8.5 b): data acquisition, mining and collection, analysis, labeling, filtration, aggregation, retention, and storage. The last item is each other operation on the data that is done before, and for, placing on the market or putting into service.

For data decommissioning, as applicable, the provider sets a mechanism for data destruction or for data archiving. The mechanism agrees with the applicable regulatory requirements. EN 18286 requires that the destruction of data does not prevent the provider from compliance with the applicable regulatory requirements.

8.6 Retirement. EN 18286 makes retirement a life cycle stage of the AI system. The standard recommends that the provider set, as applicable, the processes and tasks for the retirement of its AI systems (8.6). Examples are the risk control measures from the AI system risk management (8.2) and the data management measures (8.5). The standard requires the provider to tell the deployer about all retirement measures that are the responsibility of the deployer. Table ZA.1 has no row for clause 8.6 (see Annex ZA and presumption).

8.7 Identification of the AI system. EN 18286 requires the provider to give a "unique reference" to each AI system in the QMS scope (8.7). The provider also gives a "unique reference" to each version for which a conformity assessment procedure occurs. This reference gives identification and traceability of the AI system. The provider connects this reference to the instructions for use (8.9.3), the technical documentation (8.9.2), and other "relevant" information, for example datasets.

8.8 Continuous learning AI systems. EN 18286 clause 8.8 applies to an AI system that continues to learn after it is placed on the market or put into service. For such an AI system, EN 18286 requires the provider to do these tasks:

• Determine and document the pre-determined changes from continuous learning during design and development.
• Keep the AI system in compliance with all applicable regulatory requirements for these changes.
• Include the pre-determined changes and the behavior of the AI system in all applicable life cycle stages.
• Include all pre-determined changes in the AI system risk management.
• Verify and validate all pre-determined changes.
• Log the changes to the AI system: the pre-determined changes and the other modifications (8.8 f)).

EN 18286 also requires the provider to make sure that the technical documentation contains this information (8.8 g)):

• How the AI system learns and changes during operation
• Full information about the pre-determined changes and their effect, also on the performance of the AI system
• The data for continuous learning
• Other items of continuous learning, as applicable

The instructions for use tell deployers that the AI system learns during operation, and give the effects on the AI system (8.8 h)). As applicable, the instructions for use also give the items that are important for deployment, operation and maintenance. A NOTE to 9.4 refers pre-determined changes to clause 8.8, and clause 9.4 applies to the other modifications.

For high-risk AI systems that continue to learn, the EU AI Act gives a rule in Article 43(4), subparagraph 2. Changes that the provider pre-determined at the initial conformity assessment, and that are in the technical documentation in point 2(f) of Annex IV, are not a substantial modification:

For high-risk AI systems that continue to learn after being placed on the market or put into service, changes to the high-risk AI system and its performance that have been pre-determined by the provider at the moment of the initial conformity assessment and are part of the information contained in the technical documentation referred to in point 2(f) of Annex IV, shall not constitute a substantial modification.

8.9 Product documentation. Clause 4.5 gives the documented information for the QMS. Clause 8.9 gives the documentation that the provider prepares for each AI system (8.9.1).

For each AI system, EN 18286 requires the provider to establish and maintain technical documentation (8.9.2). The technical documentation contains full information about the AI system, its parts, and its development process, as the applicable regulatory requirements require. This information shows compliance to notified bodies and competent authorities. The EU AI Act gives its requirements for technical documentation in Article 11 and Annex IV.

The provider prepares the technical documentation before placing on the market or putting into service, and keeps it up to date. The provider also changes it when a modification of the AI system has an effect on the applicable regulatory requirements.

In Table ZA.1, the row for the first sentence of Article 11(1) of the EU AI Act includes only clause 8.9.2. The row has a condition: the technical documentation contains all the items of Annex IV.

For each AI system, the provider implements systems and procedures to establish and maintain the instructions for use (8.9.3). A NOTE to 8.9.3 refers to EN 18229-2 for more information. The Bibliography of EN 18286 gives EN 18229-2 as a standard on accuracy and robustness, under preparation (Bibliography [28]).

Clauses 9.1–9.5 — Operations and control

EN 18286 clauses 9.1 to 9.5 give requirements for the provider on identification, the supply chain, modifications, and post-market monitoring. They give recommendations on support services (9.2). For clauses 9.6 and 9.7, see Incidents and non-compliance.

9.1 Placing on the market and putting into service. EN 18286 requires the provider to have measures that give each AI system on the market or in service an identification that no other AI system has (9.1). A serial number is the example that clause 9.1 gives. The identification gives traceability of each AI system. This traceability helps record-keeping, post-market monitoring, and compliance with the other applicable regulatory requirements.

A NOTE to 9.1 tells providers that this identification is different from the "unique reference" of clause 8.7. The 9.1 identification is for the control of the distribution chain after placing on the market or putting into service.

9.2 Support services. Clause 9.2 only recommends. EN 18286 recommends that the provider identify the support services, write their specifications, and give them (9.2). EN 18286 also recommends that the provider examine these items, as applicable:

• The regulatory requirements
• The persons and organizations that support services are for, for example deployers or users
• The types of problems that can occur, and the actions that are "appropriate" for each type
• The channels for support services, and the "diagnostic tools" that get information from these channels
• A mechanism that lets all deployers send the provider the feedback that they get about possible risks to health, safety and fundamental rights

Table ZA.1 has no row for clause 9.2.

9.3 Supply chain. EN 18286 requires the provider to set and implement supply-chain control measures (9.3.1). These measures make sure that the items from external suppliers fulfill the requirements that are necessary for the AI system to comply with the applicable regulatory requirements. The items are products, components, services, model training, and test data, also for life cycle tasks. An EXAMPLE to 9.3.1 gives external design and development work, data annotation, and evaluations and tests.

Clauses 9.3.2 to 9.3.5 give these requirements:

• Selection of suppliers (9.3.2). The provider sets and documents criteria for the selection of external suppliers and for their evaluation. The criteria examine if the supplier can supply items that fulfill the requirements of the provider. The criteria are in proportion to the risk of the AI system and its intended purpose, as the risk management system determines.
• Monitoring of suppliers (9.3.3). If suppliers will continue to supply, the provider plans and documents their monitoring and their evaluation again. The monitoring plan includes criteria, frequency, and methods for the evaluation of the performance of the supplier. The provider keeps documented information about the monitoring, its results, and the necessary actions.
• Specifications for suppliers (9.3.4). The provider gives the suppliers the specifications that are necessary to fulfill the requirements of its QMS. The provider does this as applicable, and in proportion to the role of the supplier and to the type of supplied item.
• Control of the supplied items (9.3.5). The provider makes sure, and documents, that its QMS controls the supplied items. The provider sets and documents the controls on the supplier and the controls on the supplied items. The provider examines the possible effect of the supplied items on the capacity of the provider to always fulfill the requirements, the applicable standards, and the applicable regulatory requirements. The provider also examines the effectiveness of the controls of the supplier. The provider determines and documents the verification, validation, checks to accept the supplied items, or other checks.

In Table ZA.1, the row for Article 17(1)(l) of the EU AI Act includes clause 9.3, with clauses 7.1 and 7.2. The row for Article 17(1)(c) includes clause 9.3.5.

9.4 Modifications to AI systems. EN 18286 requires the provider to implement a process for the management and control of modifications to AI systems (9.4). The process includes the actions in each life cycle stage that are necessary for the applicable regulatory requirements. The provider does a review of the effects of modifications with the risk management system. After the AI system is placed on the market or put into service, the process includes a decision for each modification: substantial modification or not. The provider documents the "rationale" for each decision, for a substantial modification and for a modification that is not substantial.

Article 3(23) of the EU AI Act gives the definition of substantial modification:

‘substantial modification’ means a change to an AI system after its placing on the market or putting into service which is not foreseen or planned in the initial conformity assessment carried out by the provider and as a result of which the compliance of the AI system with the requirements set out in Chapter III, Section 2 is affected or results in a modification to the intended purpose for which the AI system has been assessed

Clause 3.2.7 of EN 18286 gives the same definition with one change. It refers to the applicable regulatory requirements (3.1.26), not to the requirements of Chapter III, Section 2 (Note 1 to 3.2.7).

In EN 18286, an AI system after a substantial modification is a new AI system, and it requires a new conformity assessment. The provider then makes the necessary changes to the QMS and applies EN 18286 to the new AI system. The technical documentation includes all versions of the AI system. When an AI system in the QMS has a substantial modification, clause 10.2.2 requires the provider to find if a review of the QMS processes is necessary (see The QMS clauses).

For high-risk AI systems, the EU AI Act gives the rule in Article 43(4), subparagraph 1:

High-risk AI systems that have already been subject to a conformity assessment procedure shall undergo a new conformity assessment procedure in the event of a substantial modification, regardless of whether the modified system is intended to be further distributed or continues to be used by the current deployer.

Subparagraph 2 tells that some pre-determined changes to a high-risk AI system that continues to learn are not a substantial modification (see 8.8 above).

The conformity assessment page gives more information about Article 43(4).

9.5 Post-market monitoring. EN 18286 requires the provider to establish and document a post-market monitoring system (9.5.1). The system agrees with all of these conditions:

• It uses a post-market monitoring plan.
• It applies from the time when each AI system in the QMS scope is placed on the market or put into service, until the retirement of that AI system.
• It lets the provider evaluate if each AI system continues to comply with the essential requirements (4.4.2).
• It is in proportion to the type of AI technologies and to the risks of the AI system. These are the risks for its intended purpose and for its reasonably foreseeable misuse.
• It gives processes to collect and review the experience from the use of AI systems. These processes find which corrective actions, or actions to prevent a problem, are necessary immediately.

The post-market monitoring plan is a part of the technical documentation (8.9.2).

The provider determines and documents an effective method for post-market monitoring (9.5.2). The method examines these items:

• The risks of the AI system, and the risk control measures from the AI system risk management (8.2)
• The applicable regulatory requirements, and the data privacy requirements
• The possible "reliance" on other organizations, for example distributors, importers, and deployers, and on third parties that supply tools, services, components or processes
• The intended purpose and the reasonably foreseeable misuse of each AI system
• The "technical constraints" that are to be "addressed" to help effective post-market monitoring
• The performance of the AI system, for the evaluation of its continuous compliance with the essential requirements
• If "relevant", interaction with other AI systems
• Monitoring and record-keeping

For effectiveness, the method is active and systematic, and it gives input to performance evaluation (clause 10).

The provider determines policies and procedures to collect and keep information from the use of each AI system (9.5.3). This is the information that is necessary for post-market monitoring. The collection is systematic. As applicable, this information comes from these sources:

• The deployer, users, or other interested parties
• The monitoring of the AI system or of its logs
• Regulatory authorities
• Feedback and complaint mechanisms, and serious incidents

As "appropriate", the provider implements AI system logging to record "relevant" data about the AI system. The provider also implements procedures for new and "emerging" hazards (9.5.4). These procedures identify the hazards from the monitoring and from the review of this information. They also start the necessary actions.

Clause 9.5.5 applies if the provider cannot, without the deployer, monitor an AI system directly and find non-compliance. Then, if it is necessary to decrease the risk, the provider puts "appropriate" requirements for monitoring in the instructions for use. As applicable, the instructions for use then include these items:

• The "technical" requirements for how to monitor the AI system, which agree with the post-market monitoring plan
• The tools that the provider recommends for this monitoring, if the AI system does not include them
• The "technical" competence that the provider recommends for the monitoring of the AI system

In Table ZA.1, the row for Article 17(1)(h) of the EU AI Act includes clauses 9.1 and 9.5. Article 17(1)(h) is about a post-market monitoring system "in accordance with Article 72". The row for Article 17(1)(k) includes clause 9.5.3. Table ZA.1 has no row for Article 72 as such.

Clause 8 and clauses 9.1–9.5 in Modulos

Modulos holds clause 8 and clauses 9.1 to 9.5 of EN 18286 in the AI System Framework MFF-19. Clause 4.3 requires the provider to determine the AI systems that the QMS includes. Modulos recommends one AI System with MFF-19 for each of them.

The AI System Requirement table gives the Requirement for each clause.

Related pages

EN 18286 overview

The publication of EN 18286, the providers that it is for, its structure, and its Modulos Frameworks

The QMS clauses

Clauses 4–7 and 10: the QMS for all AI systems of the provider

Incidents and non-compliance

Clauses 9.6 and 9.7, with Articles 20 and 73 of the EU AI Act

Annex ZA and presumption

The Table ZA.1 rows for clause 8 and clauses 9.1–9.5, with their conditions

Sources of this page

• EN 18286:2026 — Artificial intelligence — Quality management system for EU AI Act regulatory purposes, CEN/CLC/JTC 21, July 2026. CEN approved it on 12 July 2026. This page uses clauses 0.2, 3.2.7, 8.1 to 9.5, Annex ZA (Table ZA.1), and Bibliography [6] and [28]. CEN-CENELEC news of 31 July 2026.
• Regulation (EU) 2024/1689 (EU AI Act), consolidated text of 27 July 2026: Articles 3(23), 11(1), 17(1), 43(4), and Annex IV. EUR-Lex.
• The Commission Implementing Decision (EU) 2026/2274 of 12 October 2026, published in the Official Journal on 14 October 2026: Article 1 and the Annex. EUR-Lex.

Disclaimer

This page gives general information only. It is not legal advice.